## Warning: Paket 'here' wurde unter R Version 4.5.2 erstellt
## here() starts at C:/Users/jucke/OneDrive - Universitaet Bern/BSc Geographie/03 HS 25/Proseminar Geocomputation and Earth Observation/Project/toms_agds_proseminar_project
## Warning: Paket 'ggplot2' wurde unter R Version 4.5.2 erstellt
## Warning: Paket 'knitr' wurde unter R Version 4.5.2 erstellt

Introduction

Gives the background, explains the relevance, and motivates the (research) question. The introduction may also motivate the methodological approach, but doesn’t give its details.

Railway operation is affected by a range of environmental factors. In Switzerland, these effects are accentuated in autumn, when precipitation increases, temperatures fall and many trees drop their leaves. It is common knowledge that this season is prone to lower punctuality, which is a measure of quality of this mode of transport. This project investigates the influence of one of these variables, precipitation. The script is built to analyse a recent 24h timeframe.

Methods and Data

First, daily public transport data is obtained from an open data source opentransportdata.swiss. The raw data contains one data point for every ride conducted on that day. E.g. The stop of train service R16 in Biel with planned and actual arrival and departure time. Because the dataset also contains non-swiss stops and bus and ferry services, its filtered for to the relevant extent first. Then delay on arrival is calculated from planned vs. actual values. Delays on departure are ignored because it is assumed that they would not be caused by precipitation but rather from human influence. Then the train stations are assigned their geographical coordinates by matching the BPUIC number to an open source table containing their location opentransportdata.swiss. Next, the data is aggregated to show delay rate per stop and per hour. Precipitation data stems from the radar product CombiPrecip (CPC) by the Federal Office of Meteorology and Climatology MeteoSwiss. Data is provided as .h5 raster files with a time resolution of 10 minutes, containing the precipitation total of the previous 60 minutes. Processing is conducted to match the hourly punctuality data structure. Hourly precipitation total is then extracted at the exact locations of the train stations. Eventually, global and hourly correlation is investigated.

Results

Punctuality

Hourly punctuality usually shows high regional variability. A delay rate of 0 means no trains were delayed during that hour on that stop. A delay rate of 1 means all trains were delay. Selecting the stops displays the amount of trains.

## Precipitation

Hourly precipitation total is display can be explored through an interactive map. This allows for a first visual, spatial comparison of precipitation and delay concentration.

Punctuality and Precipitation Synthesis Map

Merging the two variables into one map yields the precise precipitation volumes at the stations.

Correlation

The pearson correlation between hourly precipitation and punctuality shows no significant relationship. Correlation values are comparatively low and also not consistently positive or negative for different hours of the day.

Plotting the variables on a scatterplot confirms this. The different data points, delay rate and precipitation at a specific stop, seem to show no apparent pattern.

The same applies when plotting precipitation against total delay minutes at stops instead of the punctuality rate. In the plot below, I have excluded stations with excessive amount of total delay minutes (>1000min), which likely were caused by external factors.

Analysing hourly correlation values yields no clearer picture. During the second half of the day, when precipitation increased, correlation is actually negative.

knitr::include_graphics(here("figures", "correlation_hourly.png"))

Discussion:

The results suggest that precipitation is not the single, significant factor influencing train punctuality. However, the results should be considered with caution. The analysis conducted contains some important shortcomings. Firstly, the time period of 24 hours is too short. By analysing a longer time series, e.g. 5 years, seasonal patterns might be detected. Secondly, precipitation values are extracted at the coordinates of the train stops only. Including precipitation at the immediate sourounding of a stop might improve the model. Thirdly, train delays can arise from a multitude of influences. Adding more environmental values such as type of precipitation, temperature or humidity or correcting for human influence could result in a better r2 score. Fourthly, instead of using single stop stations, analysing train services along lines (e.g. Zurich - Olten - Bern) could also improve significance of the findings.